This study presents an integrated an‎d intelligent framework for oil-production rate forecasting that combines data-driven learning, physics-constrained hybrid modeling, an‎d optimization-based enhancement to achieve higher accuracy, stability, an‎d interpretability. Traditional decline-curve models, such as Arps, Duong, an‎d Pan-CRM, offer simplicity but fail to capture nonlinear flow behavior an‎d complex reservoir heterogeneity. To overcome these limitations, the research eva‎luates advanced forecasting architectures including LSTM, GRU, DeepAR, an‎d Prophet for short-term predictions, alongside a physics-constrained BiGRU–DHNN model designed to preserve long-term physical consistency. Additionally, a modified Aquila Optimizer with Opposition-Based Learning (AOOBL) integrated with an Adaptive Neuro-Fuzzy Inference System (ANFIS) was employed to automate hyperparameter tuning an‎d enhance convergence efficiency. Comparative analyses revealed that DeepAR achieved the lowest Mean CRPS in short-term forecasts, while the BiGRU–DHNN framework sustained higher accuracy over extended horizons. The hybrid AOOBL-ANFIS model further improved robustness, achieving R² ≈ 0.95 an‎d RMSE ≈ 0.076 across datasets. Collectively, these findings demonstrate that the synergy between physics-based principles an‎d machine-learning intelligence provides a reliable, adaptive, an‎d physically interpretable foundation for next-generation digital-oilfield forecasting systems.

11/17/2025 12:00:00 AM

88784

1404/09/17

Analysis of Recent Advances in Oil Production Rate Forecasting Using Physical Models an‎d Deep Learning Approaches

Oil Production Forecasting , Machine Learning , Physics-Constrained Modeling , Hybrid Intelligence , Optimization Algorithms